Data-driven discovery of host-microbiome protein-protein interactions

Cornell University, Ithaca NY

Investigators

Abstract

Research Summary Despite numerous large-scale metagenomic case-control studies that associate microbial taxa and genes with specific diseases, surprisingly little is known about the molecular mechanisms underlying these associations. Much of the mechanistic microbiome research to date cites the production of small molecules, immune interactions and indirect interactions with food or drugs. However, direct protein-protein interactions (PPIs) have emerged as a novel facet of host-microbiome interaction underlying microbiome-associated disorders. Building from an initial analysis of host-microbiome PPIs that leveraged existing PPI databases, we have now expanded this approach to predict novel host-microbiome PPIs. We have applied protein language models to develop a novel commensal host-microbiome interaction prediction (CHIP) tool. Our final model is capable of screening through the 1011 possible host-microbiome PPIs with high precision, albeit low sensitivity. This initial step cuts down on the search space, enabling refinement steps on a much smaller subset, using structure prediction models, such as AlphaFold-multimer. This proposal aims to further develop our pipeline and validate it computationally and experimentally. Our overall goal is to accelerate data-driven and experimental discovery of host-microbiome PPIs. Our first aim is to further improve our CHIP model and analyze the host-microbiome interaction network. Our second aim is to utilize structure predictions, specifically focusing on interfaces, to infer functions. As shared interfaces between PPIs or protein-molecule interfaces suggests a common underlying function, we will examine whether microbiome proteins bind human proteins at important sites vis à vis the interactor's human protein binding partners. Our third aim is to apply proximity labeling to identify novel host receptor-bacterial protein ligand interactions, addressing a major gap in the existing data on host-microbe PPIs. Our fourth aim is to validate subsets of the host-microbiome interaction network that represent important hubs involving proteins situated within disease-relevant pathways. These include investigating the role of host-microbiome PPIs in intestinal barrier integrity, ubiquitination pathways, and inflammation signaling pathways. Overall, we aim to vastly expand our annotations of microbiome-derived proteins in human disease-relevant pathways. We anticipate these findings will lead to the identification of novel therapeutics, diagnostics and drug targets.